Evolution and Human Behavior 39 (2018) 490–501

Contents lists available at ScienceDirect

Evolution and Human Behavior

journal homepage: www.elsevier.com/locate/ens

The cognitive and cultural foundations of moral behavior T Benjamin Grant Purzyckia,*, Anne C. Pisora, Coren Apicellab, Quentin Atkinsonc,d, Emma Cohene,f, Joseph Henrichg, Richard McElreatha, Rita A. McNamarah, Ara Norenzayani, Aiyana K. Willarde, Dimitris Xygalatasj a Department of Human Behavior, Ecology, and Culture, Max Planck Institute for Evolutionary Anthropology, Germany b Department of Psychology, University of Pennsylvania, USA c Department of Psychology, University of Auckland, New Zealand d Max Planck Institute for the Science of Human History, Germany e Institute of Cognitive and Evolutionary Anthropology, University of Oxford, UK f Wadham College, University of Oxford, UK g Department of Human Evolutionary Biology, Harvard University, USA h School of Psychology, Victoria University of Wellington, New Zealand i Department of Psychology, University of British Columbia, Canada j Department of Anthropology, University of Connecticut, USA

ARTICLE INFO ABSTRACT

Keywords: Does moral culture contribute to the evolution of cooperation? Here, we examine individuals' and communities' Morality models of what it means to be good and bad and how they correspond to corollary behavior across a variety of Cross-cultural ethnography socioecological contexts. Our sample includes over 600 people from eight different field sites that include for- Cognitive anthropology agers, horticulturalists, herders, and the fully market-reliant. We first examine the universals and particulars of Evolution of cooperation explicit moral models. We then use these moral models to assess their role in the outcome of an economic experiment designed to detect systematic, dishonest rule-breaking favoritism. We show that individuals are slightly more inclined to play by the rules when their moral models include the task-relevant virtues of “honesty” and “dishonesty.” We also find that religious beliefs are better predictors of honest play than these virtues. The predictive power of these values' and beliefs' local prevalence, however, remains inconclusive. In summary, we find that religious beliefs and moral models may help promote honest behavior that may widen the breadth of human cooperation.

1. Introduction operation, few actually measure its effects directly and model it as a distributed, superordinate property of social life (see Smaldino, 2014). Many theories hold that socially learned moral norms are the Most empirical studies consider culture indirectly by either a) having lynchpin for the remarkable breadth of cooperation that humans un- participants in economic experiments make an allocation with money iquely exhibit (Bowles & Gintis, 2003; Boyd, 2018; Boyd & Richerson, and then asking what the appropriate decision was (e.g., Gurven, 2009; Chudek & Henrich, 2011; Richerson et al., 2016). However, there Zanolini, & Schniter, 2008; Ensminger & Henrich, 2014; Henrich & are a few critical outstanding issues that make this view difficult to Henrich, 2014, b) framing experimental introductions in locally salient endorse with a confidence borne out by direct empirical evidence. First, ways (e.g., Brodbeck, Kugler, Reif, & Maier, 2013; Cohn, Fehr, & it is not immediately obvious that individuals' and groups' moral pre- Maréchal, 2014; Cronk, 2007; Gerkey, 2013; Lesorogol, 2007; Stagnaro, scriptions actually influence the behavior of those who espouse them Arechar, & Rand, 2017), or c) conducting studies across multiple (e.g., Graham, Meindl, Koleva, Iyer, & Johnson, 2015; Haidt, 2001; groups, and concluding that cross-cultural variation in behavior reflects Perry, 2017; Smith, Blake, & Harris, 2013). When moral prescriptions underlying variation in culture (e.g., Apicella, Marlowe, Fowler, & and behavior are consistent with each other, moral prescriptions might Christakis, 2012; Henrich, 2000; Henrich et al., 2004; Roth, Prasnikar, simply be rationalizations of behavior rather than causes (e.g., Okuno-Fujiwara, & Zamir, 1991). Third, many cross-cultural studies Baumard, 2016; Haidt, 2001). Second, despite the fact that so many emphasizing the evolved psychology underlying morality rely heavily emphasize (or minimize) the importance of culture for human co- on theoretically-motivated scale designs (e.g., Curry, Chesters, & Van

* Corresponding author. E-mail address: [email protected] (B.G. Purzycki). https://doi.org/10.1016/j.evolhumbehav.2018.04.004 Received 12 October 2017; Received in revised form 2 March 2018; Accepted 24 April 2018 1090-5138/ © 2018 Elsevier Inc. All rights reserved. B.G. Purzycki et al. Evolution and Human Behavior 39 (2018) 490–501

Lissa, n.d.; Graham et al., 2011) that a) use items lacking in local re- 2.2. Measuring components of moral systems levance, b) are impractical for innumerate and/or nonliterate popula- tions, c) presuppose that samples have the lexical equivalent of “moral,” 2.2.1. Evolutionary psychology of morality and d) do not link this data to quantitative behavior. Contemporary evolutionary psychological research focused on Here, we seek to overcome these limitations by measuring moral mapping the conceptual space of morality typically relies on scale items culture from a variety of societies and examine whether or not moral (Curry et al., n.d.; Graham et al., 2011) with prefabricated materials values and their distributions actually have an impact on the kind of that are verified externally (i.e., using other scales). For example, broader cooperation typified by humans. We first briefly spell out our seeking to better operationalize the moral domain with attention to assumptions and introduce contemporary evolutionary perspectives on cross-cultural validity, the popular “Moral Foundations” literature moral systems, followed by a more detailed assessment of the afore- breaks down the evolutionary and cognitive “foundations” of morality mentioned limitations. We then introduce our two studies. The first into a few core dimensions. While the rubric itself has evolved (Graham consists of an analysis of systematically collected ethnographic data et al., 2013), the most recent iteration includes (1) harm/care; (2) regarding what it means to be “good” and “bad” across eight different fairness/reciprocity; (3) ingroup/loyalty; (4) authority/respect, and (5) field sites. In doing so, we examine cross-cultural moral universals and purity/sanctity as foundational to moral reasoning. The more recent local particulars. The second study uses this data to examine its con- “Morality-as-Cooperation” literature (Curry, 2016; Curry et al., n.d.) tribution to corresponding behavior in an experimental game designed measures seven types of cooperation treated as the foundations for to distinguish dishonest favoritism from impartial, rule-following fair- moral behavior: (1) family values; (2) group loyalty; (3) reciprocity; (4) ness. We conclude with a discussion of our studies' limitations and dominance; (5) deference; (6) fairness; and (7) rights to property. comment on avenues for further inquiry. These rubrics were not designed to assess the relationship between moral culture and behavior. Rather, they seek to identify variation in moral reasoning as indicated by variation in how survey items load onto 2. Background principal components and how mean values of scales vary across different groups. There are practical and methodological reasons to be reluctant to 2.1. Defining moral systems employ scale-based surveys in populations where they were not designed. First, many traditions lack the lexical equivalent of “morality.” Second, We refer to “moral models” here as the content and structure of some samples struggle with scale-based survey instruments. While con- individuals' explicit representations of moral norms. If we adopt the venient for researchers, in practice, scale items can be quite taxing and view that “culture” is shared, socially transmitted information (cf. Boyd unintuitive for non-literate and/or innumerate participants (e.g., Gurven, & Richerson, 1988; R. G. D’Andrade, 1981; Sperber, 1996), then moral Von Rueden, Massenkoff, Kaplan, & Lero Vie, 2013). Third, such instru- culture is the shared, socially transmitted units that comprise individual ments are often limited in local relevance. For example, the “Moral moral models. Defined in this fashion, local prevalence of particular Foundations Questionnaire” (Graham et al., 2011) includes questions units of socially transmitted information indicates how “cultural” or about whether or not “being good at math,” having “love for one's “normative” those units are. In this view, then, directly assessing country,” being “denied rights,” and “God's approval” are “relevant to whether or not culture influences individual behavior requires 1) de- [participants'] moral thinking” or to their sense of right and wrong. Such tailing individuals' models, 2) assessing how widespread the content of items and the notion of “moral relevance” are simply unintelligible in those models is in individuals' social groups, 3) examining the re- many contexts. Ideally, scale design in cross-cultural research begins with lationship between a behavioral trait and an individuals' models, and 4) preliminary ethnographic inquiry to ensure that scale items are actually examining the relationship between the trait and how prevalent specific measuring target constructs (Bernard, 2011; Handwerker, 2001). Indeed, informational units are in one's group. The first two requirements are Smith et al. (2007) found that other theory-driven classification schemes descriptive, ethnographic accounts of moral culture. The latter most inadequately captured the variation in folk-models of what it means to be two allow us to disambiguate the relative impacts of individual and “good” in seven different communities. Boehm (1980) imported a mor- cultural models of morality on behavior. If moral culture predicts moral ality metric to Montenegro, but due to participants' initial off-target re- behavior, then the prevalence of moral models' constituent units in a sponses to the metric, he had to assess features of local moral behavior group should covary with the target behavior. with open-ended questions. We use “moral systems” here to refer to moral models, their psy- chological underpinnings, behavioral expressions, cultural prevalence, 2.2.2. Cultural evolutionary ecology of moral behavior and the causal links between them (cf. Alexander, 1987; Haidt & Those who emphasize culture's effects on cooperative behavior ty- Kesebir, 2010; Kiper & Sosis, 2014). Classical philosophical and con- pically employ economic experimental games as an index of coopera- temporary social psychological views of moral systems emphasize tion, but do not directly measure or model “culture.” Some appeal to universality and/or the view that morality is associated with abstract the importance of cultural institutions (i.e., shared pools of norms that notions like “justice” and “rights” (Caton, 1963; Kant, 1997 [1785]; constrain human interactions in specific, socially demarcated contexts; Turiel, 1983, 2006). In contrast, many evolutionary views boil down see D’Andrade, 2006; North, 1991; Searle, 1995) by manipulating the moral systems to the regulation of cooperative and/or mutualistic en- cultural relevance of experiments' instructions in the form of framing deavors that generate individual- and/or group-level benefits effects (Brodbeck et al., 2013; Cohn et al., 2014; Cronk, 2007; (Alexander, 1987; Baumard, André, & Sperber, 2013; Barrett et al., Lesorogol, 2007; Gerkey, 2013). Others conduct experiments and infer 2016; Cosmides & Tooby, 2005; Cronk, 1994; Curry, 2016; Darwin, that culture contributes to the evolution of cooperation by virtue of 1871; Greene, 2013; Fehr, Fischbacher, & Gächter, 2002; Haidt & statistical divergences between groups in experimental game outcomes Kesebir, 2010; Machery & Mallon, 2010; Mizzoni, 2009; Sripada & (Apicella et al., 2012; Ensminger & Henrich, 2014; Henrich, 2000; Stich, 2006; Trivers, 1971; Tomasello & Vaish, 2013). However, there is Henrich et al., 2004; Roth et al., 1991). A burgeoning literature that considerable variation in moral systems, variation that many suggest actively measures variation in cultural information focuses on religious are inconsequential or run counter to such generalist theories beliefs (McNamara, Norenzayan, & Henrich, 2016; Johnson, 2016; (Baumard, 2016; Boehm, 1980; Buchtel et al., 2015; Fessler et al., 2015; Purzycki et al., 2016a). This literature typically uses individuals' beliefs Schwartz, 2007; Shweder, Much, Mahapatra, & Park, 1997; Smith, in punitive and knowledgeable deities to predict cooperative outcomes. Smith, & Christopher, 2007). As we detail below, piecing together the However, the literature ignores the within-group distribution of re- constituent parts of moral systems in a cross-cultural empirical project ligious beliefs–that is, groups' religious culture–as a factor in individual remains a major challenge in the evolutionary literature. behavior.

491 B.G. Purzycki et al. Evolution and Human Behavior 39 (2018) 490–501

To the best of our knowledge, only a solitary study claims to assess small-scale communities, whose means of living are subsistence-based the degree to which specifically moral culture has an effect in co- with daily interactions consisting primarily with local familiars as well operation (Gurven et al., 2008; cf. Ensminger & Henrich, 2014; Henrich as urban samples where individuals regularly interact with anonymous & Henrich, 2014, for more case studies from the same cross-cultural others. Moreover, our samples are uniquely poised for consideration of project). The study, which the authors characterize as “the first of its our research questions and methods. No such work examining explicit kind to show that local culture matters in explaining variation in pro- moral models has been conducted among traditional, small-scale so- social behavior” (pg. 589), employed a variation of the Dictator Game cieties. Our methods (see below) are particularly useful for eliciting rich in which participants were asked to identify “the morally correct offer and comparable ethnographic data in innumerate and nonliterate to give in this...game” (3). While there were a few exceptions in their samples often ignored or left out in studies relying on prefabricated sample of nine Tsimane’ (Bolivia) villages, in general there was a po- scales and/or narrow samples. sitive correlation between what people thought they should do in the ’ game and how they actually allocated money. Note that the Tsimane 4. Study 1: moral culture lack a word for “moral”; the researchers instead used ruijsis which “ ” expresses the concept of appropriate behavior or action (pg. 592). While not every group has the lexical equivalent of “moral,” some “ ff The authors argue that local di erences [in cooperation] are not ne- posit that the distinction between “good” and “bad” is a human uni- cessarily due to strong norms per se that vary among villages, but due to versal (Brown, 1991; Wierzbicka, 1994). We can assess whether or not ff local (unmeasured) e ects that push and pull villages towards more or the content of these conceptual domains approximate morality simply ” less pro-social sentiment (pg. 589). Note, however, that this study by asking people what it means to be good and bad. To reliably capture “ ” elicited participants' views of the appropriate behavior for the ex- moral models and culture, we assess freely-elicited data of what it periment, not more general individual-level moral models or their means to be “good” and “bad” (see Buchtel et al., 2015; Purzycki, 2011, distributions from which participants ostensibly drew. 2016; Smith et al., 2007, for precedent applications). This method al- In summary, while many argue about culture's role in the expansion lows individuals to answer on their own terms and avoids the afore- of cooperation, and many examine a variety of important factors' con- mentioned pitfalls associated with a lack of cultural relevance, the tributions to this process, the cooperation literature does not directly question of what measurement instruments actually measure, or the probe the contribution of moral culture itself. Likewise, studies that elicitation of rationalizations of behavior. have measured variation in moral culture do not link them to corre- sponding behavior. Here, we attempt to assess the relationship between 4.1. Methods individual moral models and cooperation directly, with an eye to the aspects of local culture—moral models' prevalence—that may serve as All materials were translated in local languages and back-translated inputs to individual behavior. First, we assess the degree to which there into English for corroboration and subsequent edits. To obtain reliable, are universals and particulars of moral culture by examining what naturalistic, and culturally relevant data about morality, we asked people consider “good” and “bad” (Study 1). We then examine the re- participants to spective roles of individual moral models and culture in the allocation of money using an experimental economic game designed to measure • Please list up to 5 behaviors that make someone a good/virtuous/moral honest, impartial rule-following behavior towards anonymous others person. (Study 2). • Please list up to 5 behaviors that make a bad/immoral person.

3. Participants All free-list data were translated into English and subsequently sub- mitted to Purzycki for compiling and coding. All original open-ended re- We collected data in eight different field sites (see Table 1). These sponses in English and the subsequently coded data are publicly available samples included (1) the Hadza of Tanzania; (2 and 3) inland and here https://github.com/bgpurzycki/Moral-Models-Moral-Behavior for re- coastal villagers from Tanna, Vanuatu; (4) residents of Marajó island in assessment, further recoding, and analysis. See the Supplementary data for Brazil; (5) Fijians from Yasawa island; (6) Indo-Fijians from Lovu; (7) further materials, methodological notes, and per-site English translations Tyvan residents in Kyzyl, Tyva Republic; and (8) Indo-Mauritian re- of instructions of the free-list tasks. sidents of Porte aux Piment. Our sample is notably diverse; modes of subsistence range from the foraging Hadza and horticultural inland 4.1.1. Analysis Tannese to the fully market-integrated economies of Kyzyl and Porte We analyzed the free-list data using the AnthroTools package aux Piment. (Jamieson-Lane & Purzycki, 2016; Purzycki & Jamieson-Lane, 2016) for This sample exhibits some of the considerable cross-cultural di- R(R Core Team, 2016). This package calculates the cognitive salience versity known to cultural anthropology; our participants range from the of individual free-list items and tabulates their mean salience score fully market-integrated (e.g., Mauritians and Marajó Brazilians) to (Smith's S). These scores can be calculated at the sample- and sub- subsistence foragers and horticulturalists (e.g., the Hadza and Inland sample (i.e., field site) levels. Individual item salience (i) is calculated Tanna, respectively). Our sample thus includes people from traditional, with Eq. (1):

Table 1 Descriptive features of each field site. See Purzycki et al. (2016b) and the Supplementary data for further details.

Site/sample Researcher Sampling method World religion Economy

Coastal Tanna Atkinson Cluster sampling (census) Christianity Horticulture/market Hadza Apicella Entire camps None Foraging Inland Tanna Atkinson Entire community None Horticulture Lovu, Fiji Willard Door-to-door Hinduism Market Mauritians Xygalatas Random (street) sampling Hinduism Farming/market Marajó, Brazil Cohen Random sampling (census) Christianity Market Tyva Republic Purzycki Random and chain sampling (street) Buddhism Herding/market Yasawa, Fiji McNamara Door-to-door Christianity Horticulture/market

492 B.G. Purzycki et al. Evolution and Human Behavior 39 (2018) 490–501

nk+−1 i = coding of item types by their corresponding categories in the Moral n (1) Foundations (Graham et al., 2013, 2011) and Morality-as-Cooperation (Curry, 2016; Curry et al., n.d.) literatures. If we take these most salient where n is the total number of items an individual listed, and k is the items and apply their equivalent label in the Foundations and Co- order in which an item was listed. Smith's S (Eq. (2)) is a sample's mean operation typologies, it is clear that items in the “fairness/reciprocity” value of item type (Smith, 1993; Smith, Furbee, Maynard, Quick, & domain are the most salient. This suggests a greater cultural stability for Ross, 1995; Smith & Borgatti, 1997): this “foundational” category. While “honesty” appears in the Founda- ∑ i tions typology and ranks among the items with the highest salience in S = T N (2) the present free-lists, there is not a broad consensus about honesty in the literature; others simply include honesty as another moral sub- Here, we denote item type with i and N denotes the total sample or T domain (Ashton, Lee, & de Vries, 2014; Blasi, 1980; Hofmann, sub-sample size (i.e., the denominator is not limited only to those Wisneski, Brandt, & Skitka, 2014) of the greater repertoire of moral participants who listed a given item). Smith's S will therefore increase foundations and the Morality-as-Cooperation literature bypasses it. as a function of ubiquity and earlier placement in lists. In order to Regardless, these results are consistent with what we would expect minimize inflated Smith's S values due to repeated items within lists, we given the evolutionary literature's view of morality as a system reg- used AnthroTools' “MAX” function which includes only the earliest- ulating social exchange and cooperation (Alexander, 1987; Curry et al., listed repeated item in its calculations. n.d.): fairness and reciprocity loom large in mental models of what it It is important to note that our reported salience indices may reflect means to be good. “Loving” and being kind are also included. While the underestimations for three reasons. First, in terms of individual item Foundations literature considers this as part of the Care/Harm foun- salience, participants were encouraged to list up to only 5 items per dation, it is not immediately clear how these fall within the scope of the sub-domain (i.e., “good” and “bad”). If these items would have been the Cooperation typology, a limitation acknowledged in Curry et al. (n.d.). earliest-listed in a task without such a constraint, all data considered What makes an “immoral” or “bad” person? In this sub-domain, here would have had much higher salience scores. Secondly, we retain participants (N = 650) listed 2728 items (M = 4.20,SD = 1.14). items listed by only one individual in our analysis. Dropping such Table 3 details the top five items where S ≥ 0.10. Fig. 1 illustrates the idiosyncratic items would decrease the denominator in Eq. (1) and content of these models by salience. Much like the “good” data, many therefore increase the salience values. Thirdly, for the sake of com- participants reiterated the question in their responses (e.g., bad people pleteness, we include those individuals who simply answered “I don’t exhibit “bad behavior”). Also consistent with the evolutionary litera- know” (1 participant for the “good” list and 3 for the “bad”). Though ture, “theft” is the most salient item listed across cultural groups (the negligible, these would inevitably have an item salience of 1 and con- antithesis of generosity), followed by “deceit” and “violence.” Nota- tribute to a larger denominator in Eq. (2). bly—and not often considered in much of the evolutionary literature (cf. Kurzban, Dukes, & Weeden, 2010; Rozin, 1999)—the use and abuse 4.2. Results of drugs, alcohol, and other substances are among the chief items listed in what makes a “bad” person. It may be that people really do see the 4.2.1. Moral universals consumption of drugs and alcohol as bad in and of itself, but they also What constitutes a “moral” or “good” person? Participants (N may view intoxicants as the source of bad behavior and therefore in- = 643) listed a total of 2478 items (MListed = 4.27,SD = 1.07) in this dicative of immoral conduct. Here too, the Foundations literature has a sub-domain. Table 2 and Fig. 1 show the salience of individually-listed little more coverage than the Morality-as-Cooperation literature. items where S ≥ 0.10. We use this cut-off to minimize table lengths. Cross-culturally, the most salient components of individuals' mental Many participants listed various items that simply re-expressed the models of morality revolve around the provisioning of material re- question (e.g., good people have “good hearts” or exhibit “good beha- sources in the form of generosity, helpfulness, and theft. We might in- vior”). For the purposes of analysis, these items were given the same terpret honesty and dishonesty as facets of material goals as well insofar code. After this, the most salient item for participants was “generosity” as it is virtuous to be honest about how much others stand to gain or or “sharing,” followed by “helpfulness” and “honesty.” lose in interactions (per Ashton et al., 2014; Fischbacher & Föllmi- For the sake of reference and discussion, we include a post hoc Heusi, 2013, see below). However, as is made clearer in our examina- tion of site-specific moral models, there is also some variation that is Table 2 not immediately related to cooperation. Sample salience scores ≥0.10 for what makes a good person (N = 643). Salience (M) is the average individual item salience among individuals who 4.2.2. Moral particulars listed a given item, Smith's S is the individual item salience of the sample, and n While we presently have neither the room to contextualize each is the number of participants listing the item. Foundation column indicates site's specific results nor the means to make confident inferences about corresponding type in the “Moral Foundations” literature, while Cooperation type column denotes corresponding type in the “Morality-as-Cooperation” lit- why cross-cultural variation exists (i.e., we only have 8 groups and erature. Question marks indicate possible interpretations or lack of obvious cross-sectional data), the analysis of moral particulars does indicate correspondence. * denotes clustered items such as good conscience, good be- that patterns in the global sample are not merely artifacts of a few havior, good nature, good heart. groups driving the results. For the sake of concision, we present the most salient item listed by site. Table 4 details the items with the Item Foundation Cooperation type Salience (M) Smith's S n highest salience scores by site for the “good” free-list data. Table 5 Good* –– 0.73 0.22 195 details the items with the highest salience scores by site for the “bad” Generous/ Fairness/ Reciprocity 0.64 0.18 178 list. More exhaustive tables are available in the Supplementary mate- shares reciprocity rials (Tables S2 and S3). Helpful Fairness/ Reciprocity 0.59 0.17 183 reciprocity Across the most salient items of what constitutes good people, we Honest Honesty/ ? 0.69 0.14 129 find some curious idiosyncracies. For instance, Indo-Mauritians listed deception “speaks well” (i.e., polite and considerate as opposed to rude and ob- Respectful Authority/ Group loyalty 0.56 0.13 144 scene) earlier and more often in lists than did other groups. Notably, respect Indo-Fijians (Lovu, Fiji) likewise ranked “speaks well” highly Loving Care/harm ? 0.61 0.11 113 Kind Care/harm ? 0.73 0.10 88 (S = 0.10) while all other sites ranked it lower when listed at all (S values were ≤0.04). This suggests a distinct cultural lineage; both are

493 B.G. Purzycki et al. Evolution and Human Behavior 39 (2018) 490–501

Fig. 1. Universal moral models. Center circle represents the domain. The most salient items in these domains are on top with item connection weights (Smith's S values) descending clockwise.

Table 3 This term ignorante is locally nuanced, and refers to one who ignores Sample salience scores ≥0.10 for what makes a bad person (N = 650). Salience others' opinions and holds their own as superior. This is also not ob- (M) is the average individual item salience among individuals who listed a viously a component of the Moral Foundations rubric. Note that for half given item, Smith's S is the individual item salience of the sample, and n is the of the sites, “theft” had the highest salience for what constitutes a bad number of participants listing the item. Foundation column indicates corre- person; for the other half, save Marajó, group-level salience for theft “ ” sponding type in the Moral Foundations literature, while Cooperation type was > 0.10: Hadza (S = 0.31); Mauritius (S = 0.15); Tyva Republic column denotes corresponding type in the “Morality-as-Cooperation” literature. (S = 0.13). Again, we refer readers to the more thorough tables in the Question marks indicate possible interpretations or lack of obvious correspon- Supplementary materials. dence. * denotes clustered items such as bad conscience, bad behavior, bad nature, bad heart.

Item Foundation Cooperation type Salience (M) Smith's S n 4.2.3. Discussion Cross-culturally, the most salient indicators of good people are Theft Fairness/ Property rights 0.71 0.26 235 generosity and sharing, helpfulness, and honesty. The most salient in- reciprocity Dishonest Honesty/ ? 0.66 0.16 156 dicators of bad people are theft, dishonesty, and violence. Theft is no- deception table insofar as it might be construed as the antithesis of cooperation; Violent Care/harm ? 0.65 0.15 153 Baumard (2016) characterizes it as “violating the logic of balanced Drugs/alcohol Purity/ ? 0.70 0.12 108 interests” (126). These items largely correspond to rubrics offered by sanctity (?) the Foundations and Cooperation literatures, but a closer look at site- Bad* –– 0.71 0.11 101 specific moral models appear to complicate those rubrics insofar as articulateness, religious piety, drug and alcohol use and abuse, and Indian diaspora populations that may have common-source value sys- ignorance and arrogance are not consistently considered. Aside from tems. Going to church was the most salient item for Fijians from illustrating their limited breadth and narrow focus, our data pose little Yasawa. This particular value does not obviously fit in the Moral problem for the approaches; we are explicitly measuring representa- Foundations or Morality-as-Cooperation rubrics and was not ranked tional models of morality rather than moral reasoning or judgment of a highly at any other site. We nevertheless suggest “foundations” or particular behavior. That said, it does suggest that the methods used in “cooperation type” into which these might be classified in Tables 4 and this literature have not captured some potentially informative variation 5. Likewise, in Marajó, “ignorance/arrogance” has the highest salience, found in cultural representations of what it means to be (im)moral. although the small Smith's S suggests minimal consensus at this site. While the above results map cross-cultural models of morality for eight populations, one remaining question is whether or not the content

Table 4 Per-site items with highest salience scores for what makes a good person. Salience (M) is the average individual item salience within individuals who listed a given item and Smith's S is the individual item salience of the sub-sample. Sample size (n) is site-specific sample size for sub-domain. Foundation column indicates corresponding type in the “Moral Foundations” literature, while Cooperation type column denotes corresponding type in the “Morality-as-Cooperation” literature. Question marks indicate possible interpretations or lack of obvious correspondence.

Site Item Foundation Cooperation type Salience (M) Smith's S n

Coastal Tanna Generous/shares Fairness/reciprocity Reciprocity 0.74 0.40 44 Hadza Lovinga Care/harm ? 0.62 0.40 69 Inland Tanna Hospitable Care/harm Group loyalty (?) 0.74 0.45 74 Lovu, Fiji Honest Honesty/deception ? 0.73 0.38 79 Mauritius Speaks well Authority/respect (?) Deference (?) 0.84 0.33 82 Marajó, Brazil Helpful Fairness/reciprocity Reciprocity 0.70 0.18 77 Tyva Republic Honesta Honesty/deception Reciprocity (?) 0.74 0.28 115 Yasawa, Fiji Goes to church Purity/sanctity (?) Group loyalty (?) 0.68 0.39 103

a Second highest salience within sample after the lumped “good” category (see above).

494 B.G. Purzycki et al. Evolution and Human Behavior 39 (2018) 490–501

Table 5 Per-site items with highest salience scores for what makes a bad person. Salience (M) is the average individual item salience within individuals who listed a given item and Smith's S is the individual item salience of the sub-sample. Sample size (n) is site-specific sample size for sub-domain. Foundation column indicates corresponding type in the “Moral Foundations” literature, while Cooperation type column denotes corresponding type in the “Morality-as-Cooperation” literature. Question marks indicate possible interpretations or lack of obvious correspondence.

Site Item Foundation Cooperation type Salience (M) Smith's S n

Coastal Tanna Theft Fairness/reciprocity Property rights 0.76 0.46 44 Hadza Murder Care/harm ? 0.52 0.34 69 Inland Tanna Theft Fairness/reciprocity Property rights 0.76 0.37 78 Lovu, Fiji Theft Fairness/reciprocity Property rights 0.82 0.36 79 Mauritius Violence Care/harm ? 0.65 0.30 84 Marajó Ignorance/arrogance Authority/respect (?) Deference (?) 0.69 0.14 76 Tyva Republic Dishonest Honesty/deception ? 0.70 0.35 117 Yasawa, Fiji Theft Fairness/reciprocity Property rights 0.80 0.33 103 and ubiquity of moral models affect behavior. To assess this, we ex- each site except for the Hadza who played with tokens each worth 8 oz. amined the effects of the presence and cognitive salience of honesty or (∼226.80 g) maize. All participants were tested for game comprehen- dishonesty on a behavioral economic experiment that measured sys- sion and knew that all coins would be distributed to those designated by tematic and partial allocations. We predicted that in a game that the cups, including themselves. Only participants who passed the measures dishonest favoritism indicative of systematic rule-breaking, comprehension questions are included in this data set. the cognitive salience of “honesty” and “dishonesty” should predict After experiments, participants answered a host of interview ques- fairer play. In other words, by measuring whether an individual's moral tions, including the aforementioned free-list tasks. Note that in ex- model includes honesty and dishonesty, we can assess his or her re- periments were predominantly those who completed free-list tasks, but sistance to the opportunity to cheat. To assess the impact of moral there a few who did not complete free-lists or cases where individuals culture, we also modeled group-level cultural prevalence of these who participated in free-list tasks did not participate in experiments. components. They were also asked what they thought the experiment was about; their open-ended responses were coded for whether they mentioned cheating, fairness, or honesty. Participation took a total of ∼90 min, 5. Study 2: moral behavior with the free-list task typically taking place ∼15 min after the game. Briefly, there are at least five reasons why this ordering had no effect on 5.1. Methods free-list outcomes: 1) free-list tasks were after demographic surveys, 2) the game check question is not correlated with listing (dis)honesty, 3) 5.1.1. Economic experiment some sites simply did not list (dis)honesty frequently, 4) previous re- To measure honest behavior, we had participants play an economic search (Smith et al., 2007) not using experiments in this fashion shows game designed to detect dishonest favoritism (Cohn et al., 2014; that listing “honest” is quite prevalent cross-culturally, and 5) com- Fischbacher & Föllmi-Heusi, 2013; Hruschka et al., 2014; Jiang, 2013; paring the data from one site with and without the experiment shows McNamara et al., 2016; Purzycki et al., 2016a). In this experiment, no indication of games having an effect (see Section 3.4 of the Sup- participants have a stack of 30 coins, a fair 6-sided, 2-colored die, and plementary data for more details). two cups designated for a specific individual. They think of which cup All methods, materials, and data are available online at https:// they want to put a coin into and then they roll the die. If one pre- github.com/bgpurzycki/Moral-Models-Moral-Behavior. designated color appears, they are supposed to put the coin into the cup they thought of. If the other color appears, they are supposed to put the coin into the cup opposite the one they thought of. They repeat this 5.1.2. Does moral culture matter? until all 30 coins are in cups. They make these decisions alone without If moral models contribute to the expansion of cooperation, listing any outside observers. Regardless of individual decisions, coins should (dis)honesty should predict playing by the rules. Participants who be randomly allocated to either cup if participants follow instructions. mention (dis)honesty will be more likely to allocate coins to the cup However, since participants play alone, they can allocate more coins to benefiting the recipient more socially distant from themselves—that is, the cup of their preference. someone non-local (as opposed to someone local), or someone other In our study, participants played two counterbalanced games, each than the participant him- or herself. If the cognitive accessibility of task- with two cup dyads. The “Local Community Game” included one cup relevant components of moral models is also important to the expansion reserved for an anonymous co-ethnic, co-religionist in the participant's of cooperation, we should see that salience of listing (dis)honesty in- local community and one was for an anonymous, geographically distant creasing the chances of allocating coins as well. If moral culture con- co-religionist who was also a co-ethnic by default. In the “Self Game,” tributes to the expansion of cooperation, then within-sample ubiquity of one cup was reserved for the player and the other cup for another (dis)honesty in moral models should also have an effect on individual anonymous co-ethnic, co-religious individual in the same specified behavior. In other words, while an individual's moral model may induce geographically distant region. Players got to keep the money that went fairer behavior, living in a context where more people share similar into their own cups and we distributed all allocations to randomly-se- moral values–along with the expected repercussions of violating those lected individuals designated by the other cups. Here, the geo- values–should also contribute to the likelihood of playing honestly. graphically distant players function as an index of the kind of broader Conversely, if moral culture evolves in response to local problems, it cooperation that is unique to humans; as participants are not likely to may actually be associated with more self-interested behavior. Previous ever interact with these distant players, playing according to the rules work we build upon shows that individual-level beliefs about morally indicates an unwillingness to favor themselves or their community. concerned gods' punishment and knowledge breadth (i.e., omniscience) Show-up fees for participation were ∼25% of the average daily predicted allocations to the distant play (Purzycki et al., 2016a). By the wage in our field sites. We set aggregate stakes at roughly a single day's same logic, then, the more one's community claims that morally con- wage (x) where individual die rolls were worth the closest coin in value cerned deities know and punish people, the more likely individuals to x/number of games played/30 coins. Coins were real currency in should behave fairly (or not). In sum, in addition to individuals' moral

495 B.G. Purzycki et al. Evolution and Human Behavior 39 (2018) 490–501

Table 6 than hold intercultural variation constant, we incorporate it into our Summary statistics for individuals per site who participated in the experiment modeling structure (Gelman, 2006; Nezlek, 2010; Pinheiro & Bates, and listed “honest” and “dishonest” in the good and bad list respectively. N 2000). Third, we formally develop statistical models (see the Supple- “ ” “ denotes number in sub-sample, n denotes those who listed honest and dis- mentary data) in a Bayesian framework. Fourth, using prior defined ” honest, and Prop. refers to proportion of sub-sample listing these items. Means distributions, we impute our missing data. (M) and standard deviations (SD) are of item salience. Our outcome variable is the binomially distributed count of allo- Honesty in good list Dishonesty in bad list cating coins to the socially distant cup. As mentioned above, this offers the strongest test of fairer play as the chances of allocating a coin to Site Nn Prop. M(SD) n Prop. M(SD) geographically distant people better approximates to playing fairly than Coastal Tanna 44 7 0.16 0.75 (0.20) 6 0.14 0.69 (0.22) do allocations to self- or local community (Hruschka et al., 2014; Hadza 68 8 0.12 0.52 (0.29) 10 0.15 0.59 (0.29) Purzycki et al., 2016a). The log-odds of each allocation is defined as a a Inland Tanna 76 1 0.01 0.25 (–) 2 0.03 0.71 (0.06) linear combination of Lovu, Fiji 76 40 0.53 0.73 (0.31) 41 0.54 0.65 (0.26) Mauritius 91 4 0.04 0.57 (0.34) 4 0.04 0.55 (0.34) Marajó 77 16 0.21 0.65 (0.31) 11 0.14 0.51 (0.35) 1. Varying intercepts for individual and group Tyva Republic 79 24 0.30 0.83 (0.20) 38 0.48 0.74 (0.28) 2. Varying slopes by group for individual-level responses to moral Yasawa, Fiji 72 4 0.06 0.40 (0.16) 18 0.25 0.72 (0.26) models, moralistic gods' punishment, and moralistic gods' knowl- Total 583 104 0.18 0.70 (0.29) 130 0.22 0.67 (0.28) edge breadth 3. Fixed slopes by group for group-level average responses to moral a Note that this value reflects the salience score of the single individual. models, moralistic gods' punishment, and moralistic gods' knowl- edge breadth models and religious beliefs, within-sample ubiquity of (a) (dis)honesty 4. Simple effects for religious prime condition, order, game under- in moral models, as well as (b) beliefs about moralistic gods' omnis- standing check, game type, and number of children cience and (c) punishment, should predict fair play above and beyond the content of an. The group-level responses are given their own statistical models, as To test whether moral models affect game play, we coded whether they are not observed but rather must be inferred from the individual or not participants listed honesty or dishonesty in their free-lists and responses at each site. Rather than use simple fixed indices of group- used the summation of these two indices as a predictor with possible level variation (e.g., the mean value for the cultural variables in each values of 0, 1, and 2. As such, participants who did not answer free-list site), we infer them from the sample of individual statements. Then we tasks or answered with “I don’t know” are not considered in this ana- simultaneously use the posterior distribution of each in the model. This lysis. Note that the odds of mentioning honesty and dishonesty are retains all uncertainty so that we do not limit ourselves to point esti- related; a logistic regression shows that the odds of listing dishonesty mates that may lead to misleadingly false precision where there is ac- increase by 7.10 [95% CI = 4.48, 11.25] when participants list honesty. tually a distribution of values within communities. For each set of be- liefs—moral models, moralistic gods' punishment and knowledge 5.1.3. Participants breadth—we simultaneously estimate a varying intercept representing Table 6 reports the summary statistics for experimental participants the average of each site's individual responses and use this intercept, “ “ “ ” who listed honesty in the good list (n = 104, 51 women, mean with all associated uncertainty, as a predictor in the main model. This is “ ” age = 37.26) or dishonesty in the bad list (n = 130, 49 women, mean analogous to a measurement error model, in which the group-level age = 39.32). Note the considerable variability in the number of par- predictors are measured with error. In principle, then, our model is four ticipants who listed honesty or dishonesty across these eight popula- simultaneous regressions: a main binomial regression predicting coin “ ” tions. Ten individuals listed multiple items coded as honest for the assignments and three varying intercept regressions predicting in- “ ” “ ” good list and 9 individuals listed two items coded as dishonest for dividual responses by field site. Formal details of the model are in- “ ” the bad list. Again, these individuals are treated as listing each only cluded in the Supplementary materials. once. With one exception, each individual played both games. We there- fore restructured the data set to include two duplicate participant-by- 5.1.4. Model variable matrices and included a binary variable denoting which game Again, elsewhere (Purzycki et al., 2016a, 2016b), we found that it represents (“Self Game” = 1). We included all values for coins to individuals' beliefs about morally concerned deities' punishment and distant co-religionists in a single vector and the cups designated for the knowledge breadth contributed to fairer play in a wide variety of model local co-ethnic and participants in another single vector. The Hadza specifications. We also found that the more children people had, the were not asked the game understanding check question, so we mar- more likely they were to allocate more coins to themselves and local ginalized over these and imputed other missing values (see script for communities. As some participants in the present study played in a details, and the Supplementary data for alternate imputation strate- treatment condition using various religious primes (with no overall gies). effects detected), we hold this and game order (Local Community Game We fit this model using the R package rethinking (version 1.71) first = 1) constant in the regressions (see the Supplementary data for (McElreath, 2016, 2017) and rstan version 2.17.3 (Stan Development further details). In order to hold constant any effects for recognizing Team, 2017). We assessed chain convergence by inspecting traceplots, what the game was about, we created an indicator variable where va- R values, and the number of effective samples, and encountered no lues of 1 denote when participants thought the experiment was about problems in sampling. The Supplementary materials include further fairness, honesty, and/or cheating (n = 31; 5% of the sample). analyses in a frequentist statistical framework and analyses considering Here, we build upon the “Reduced models” in Purzycki et al. (2018). the effects of item salience. These reduced models were the result of backward-selected full models, had the lowest variance inflation factors, and largest sample sizes of any 5.2. Results other model specification. We develop these models and their appli- cation in a few important ways. First, we incorporate moral models and 5.2.1. Individual-level effects culture as predictor variables at individual and group levels, respec- Table 7 reports the main models. Values are exponentiated mean tively. Here, the group level refers to within-sample ubiquity. Second, estimates (OR) and 95% credibility intervals (CI). We have highlighted we take group-level variation into account using varying effects. Rather the individual-level effects in Fig. 2. As indicated by the intercept,

496 B.G. Purzycki et al. Evolution and Human Behavior 39 (2018) 490–501

Table 7 Exponentiated mean estimates (OR) and 95% credibility intervals (CI) of chances of allocating a coin to geographically distant co-religionists. The left-most model is the full model, the center model removes all moral variables, and the right-most model includes only individual-level moral models.

OR [95% CI] OR [95% CI] OR [95% CI]

(Dis)honesty summation 1.03 [0.96, 1.12] – 1.03 [0.97, 1.09] Moralistic gods' punishment 1.09 [0.93, 1.29] 1.10 [0.95, 1.26] 1.10 [0.96, 1.25] Moralistic gods' knowledge 1.25 [1.03, 1.53] 1.25 [1.04, 1.51] 1.23 [1.02, 1.51] Number of children 0.88 [0.70, 1.08] 0.86 [0.70, 1.06] 0.87 [0.70, 1.08] Condition (treatment = 1) 0.97 [0.90, 1.05] 0.97 [0.89, 1.04] 0.96 [0.89, 1.04] Local game played first = 1 1.02 [0.96, 1.09] 1.02 [0.96, 1.09] 1.02 [0.95, 1.10] Game about honesty? (yes = 1) 1.00 [0.85, 1.20] 1.02 [0.87, 1.22] 1.02 [0.86, 1.20] Game (self game = 1) 0.97 [0.93, 1.02] 0.97 [0.93, 1.02] 0.97 [0.94, 1.01] Group-level moral models 1.30 [0.51, 3.59] –– Group-level gods' punishment 1.65 [0.34, 6.25] 2.15 [0.59, 6.46] 2.23 [0.59, 6.55] Group-level gods' knowledge 1.10 [0.29, 4.13] 1.18 [0.36, 4.01] 1.13 [0.38, 3.41]

chances of allocating a coin to the geographically distant players. Note here that the strongest effect is gods' attributed knowledge breadth; the more individuals claim gods know, the more likely they are to allocate to the distant cup. Gods' punishment also has an effect in the same direction, though not as obviously strong. Previous results (Purzycki et al., 2016a) indicated that punishment predicted larger allocations than knowledge. However, these previous models treated field site as a simple effect and did not allow any variables to have differential effects across sites. Moreover, they considered only complete cases. Here, we allow these factors to have differential effects across sites while esti- mating the effects of individual-level factors on fair behavior. The content of moral models influences game play; individuals who listed (dis)honesty are more likely to play fairly (i.e., there is a 3% greater chance of allocating a coin to distant players). Note, however, Fig. 2. Exponentiated 95% credibility intervals of mean estimates of individual- while the bulk of the probability mass is > 1, this effect is notably slight ff level e ects of full model from Table 7. Horizontal axis is on a logarithmic scale. by comparison to religious beliefs. As indicated by the relatively nar- ff The dotted vertical line indicates the threshold of no e ect where variables with rower intervals, it is, however, better estimated by the model. In the no reliable effects would have error symmetry around 1.0. Effects to the right of Supplementary data, among a variety of other model specifications, we 1.0 predict greater odds of allocating a coin to geographically distant players whereas effects to the left indicate decreased odds in such allocations. show that salience of (dis)honesty has a similar relationship to beha- vioral outcome; individuals who list (dis)honesty earlier in lists are more likely to allocate coins to the distant players. individuals predictably bias allocations in favor of themselves and their local community across the eight sampled populations. As suggested by ff the Game variable, people favor themselves slightly more than they 5.2.2. Group-level e ects favor their local communities (i.e., the odds and range of the effect is Table 7 also includes the average contribution of moral and re- trending towards values < 1.00). Moreover, the effect of the number of ligious culture on allocations. Fig. 3 illustrates a projection of these ff children people have is trending towards favoritism for the players group-level e ects, assuming participants have no children and an- themselves and their local communities. swered all questions at the half-way mark (in this case 0.5). This in- Our focal individual-level variables—moral models, deities' pun- cludes the free-list summations, which are inverse logit transformed to ishment and knowledge breadth—all contribute to increasing the put them on the same scale as the other cultural variables. Group-level beliefs in moralistic gods' punishment (OR = 1.69, 95% CI = [0.39,

Fig. 3. Projected effects of cultural prevalence of (dis)honesty (green), gods' punishment (red), and gods' knowledge (blue) on probability of allocating a coin to distant players. Results are from full model in Table 7. Shading is 95% percentile intervals. Reference line is at the 50% mark indicating fair play. Across projections, number of children are held at zero, and all other values are held at 0.5 (the halfway point for each variable, including the inverse logit transformed values for moral models), varying only the group level values between 0 and 1. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

497 B.G. Purzycki et al. Evolution and Human Behavior 39 (2018) 490–501

6.68]) exhibits the clearest positive trend (red in Fig. 3). In other words, present study is the cognitive salience of (dis)honesty. In the Supple- the ubiquity of beliefs that gods punish in an individual's community mentary data, we show that the effects for individual-level salience of positively predicts his or her fair behavior in the game (albeit with a (dis)honesty are similar to those we find above; the earlier individuals range of error). Cultural ubiquity of (dis)honesty in moral models list (dis)honesty, the greater the odds of allocating a coin to the distant (green in Fig. 3) and gods' knowledge breadth (blue in Fig. 3) show no participant (see the Supplementary data). In addition to accessibility, reliable effect on allocations (i.e., their credibility intervals are more item salience may indicate how readily individuals can implement symmetrical around 1.00). these values. This is consistent with the psychological literature that The strongest conclusion to be drawn from the main results is emphasizes the impact of quick moral intuitions over slow moral that—as indicated by the wide intervals—these are poorly estimated judgments (Baumard, 2016; Cone & Rand, 2014; Haidt, 2001; Lotz, factors. While the odds ratios (exponentiated mean estimates) appear to 2015), suggesting that deeper motivational forces are at work. be high, the 95% credibility interval width is quite broad and it is Fair play had a stronger association with religious beliefs than with difficult to conclude that culture–the prevalence of certain kinds of moral models. The effect of being watched and potentially punished by information within a community–has a systematic effect on individual a transcendent being may have played a larger role in our expanded behavior cross-culturally. Using more liberal-but-standard analyses, sociality than moral cultivation. It also raises questions as to whether or however, we cautiously show that for populations where (dis)honesty is not human moralistic punishment is powerful enough to offset the costs infrequently listed, models predict that increasing its cultural pre- of defecting on social expectations (cf. Dawkins, 2016; De Waal, 2013; valence brings allocations to distant players in this sites to baseline, Johnson, 2016). In other words, not only moral models but additional cross-site allocation levels (Section 5.2.2 of the Supplementary data, institutional and environmental factors are likely required to stabilize Fig. S1). While individually held cultural information predicts in- wider, more predictable cooperation. The secularization literature (e.g., dividual outcomes, it remains less clear as to how cultural prevalence of Norris & Inglehart, 2012) and some of the aforementioned experimental that information does. work (e.g., Cronk, 2007; Gerkey, 2013) suggest that this is the case.

6. Discussion 6.2. Moral behavior

The studies presented here give new insight into the relationship Some argue that the evolutionary function of moral behavior is to between morality and the kind of broad cooperation unique to humans. maintain individuals' reputations in reciprocal interactions (Baumard, Among our diverse samples, the most salient and ubiquitous compo- 2016; Baumard et al., 2013; Sperber & Baumard, 2012). Our result that nents of moral culture revolve around reciprocity, cooperation, hon- gods' knowledge has the strongest and most reliable association with esty, and dishonesty. This cross-cultural ethnographic data empirically giving more coins to distant players is consistent with this view insofar confirm that moral culture is more associated with the costs and ben- as one's reputation matters in the eyes of a god. Beliefs about morally efits of social life (Alexander, 1987; Fessler et al., 2015; Greene, 2013; concerned deities that know about and punish people for immoral be- Tomasello & Vaish, 2013; Trivers, 1971) than with concerns of “justice” havior might predict moral behavior more reliably because they har- and “rights” (Turiel, 1983, 2006). We also found that individuals' moral ness–among other things–psychological systems responsible for re- models predict honest behavior towards geographically distant in- putation management and punishment avoidance (Johnson, 2016; dividuals, but their effects were not as strong as the effects of religious Purzycki et al., 2016a). If concern of one's reputation in his or her beliefs. As participants exhibited this behavior towards individuals they community was strong enough to overcome both the anonymity af- would never likely meet, our results also confirm the important role forded by these experiments and that target recipients were in no po- that individuals' beliefs and values have on human sociality by re- sition to reciprocate, we should have seen a relationship between moral straining selfish behavior. Group-level moral and religious culture, culture and honest allocations. That is, individuals should have been however, are not clearly associated with individuals' moral behavior. more likely to play honestly if it meant breaching widely held values Below, we discuss various facets of the moral system in light of these (Baumard, 2016, p. 131, n. 13). We found no such association. It may results. be, however, that moral culture functions in ways not captured by such games (see below). 6.1. Morality in mind Ongoing concerns revolve around the ecological validity of eco- nomic experiments (Baumard & Sperber, 2010; Gervais, 2017; Gurven While we found an individual-level effect of moral models on & Winking, 2008; Wiessner, 2009; Winking & Mizer, 2013). In small- honest, rule-following behavior, the effect itself was quite small. This scale societies, the anonymity afforded by these experiments is not al- may have been due to the kind of data our methods elicit; open-ended ways available. Our study takes advantage of the rarely-offered anon- questions require categorization for analysis, which may have in- ymity by examining whether or not participants exploit the experi- troduced bias. However, field researchers collected and translated our mental context for their own gains. Moreover, recall that target ethnographic data which was checked multiple times for quality and recipients were geographically distant individuals with whom partici- consistency across research assistants. Moreover, our models exhibited pants are unlikely to interact; we assessed whether or not cultural considerable precision in estimating (dis)honesty's effect on behavior. content can induce impartial and honest behavior in interactions we We also considered other notions beyond (dis)honesty that we might know are not happening regularly in our study sites. Other options for assume to be task-relevant. However, moral model components such as assessing interactions are likely to miss these rare encounters (Pisor & “cheating,”“disobedience,” or “fairness” were either concentrated in a Gurven, 2016). As our religious belief measures showed a relatively few communities or rarely listed at all, thus making it difficult to re- strong association with allocation, this suggests that we cannot easily liably assess their effect on behavior in a global cross-cultural study dismiss the value of using such games in toto. such as ours. Using the data presented here to design scales for mea- suring individuals' moral models would strike a balance between uni- 6.3. Moral culture versal applicability and local relevance (e.g., Boehm, 1979, 1980; Buchtel et al., 2015). Consistent with the precedent study examining cultural variation Moral models might be only as effective to the extent that an in- (Gurven et al., 2008), our results are inconclusive as to whether or not dividual can implement them. Accordingly, measuring other critical moral culture actually corresponds to individual behavior. As indicated individual-level factors (e.g., self-control; Blasi, 1980) to predict moral by the wide credible intervals of our model estimates, group-level moral behavior might be also appropriate. One indication of this in the or religious commitments within a community do not reliably predict

498 B.G. Purzycki et al. Evolution and Human Behavior 39 (2018) 490–501 individual behavior. It may be the case that there are unmeasured, Research ethics contextual factors that may be responsible for this model uncertainty. For instance, we do not know the relationship between moral culture This project was initially approved by the University of British and the threat of punishment for moral violations in each sample (Boyd Columbia's Behavioural Research Ethics Board (#H13-00671) and & Richerson, 1992). We also do not have a reliable sense of moral subsequently approved by the ethical review boards at the home uni- culture's relationship with socioecological factors such as material se- versity of each researcher who collected the data. curity (Hruschka et al., 2014) or environmental harshness (Gelfand et al., 2011). As our sample is limited to eight field sites, more attention Competing interests to group-level measures such as these in a larger sample would facilitate a more reliable assessment of these factors' relative contributions. We declare that we have no competing interests. However, it may be the case that behaviors that correspond to moral virtues occur too context-specifically or situationally to be reliably Acknowledgments evoked in experimental games (Fessler et al., 2015; Gerkey, 2013; North, 1991). We do not have a precise grasp of the components of This work was made possible by the Cultural Evolution of Religion individuals' moral models that become salient when they operate in Research Consortium, funded by a SSHRC partnership grant (grant different contexts (or whether or not they do). At the group level, rather number 895-2011-1009) and the John Templeton Foundation (grant ID than having a direct, measurable effect on our behavior, cultural ubi- 40603) awarded to J.H. and A.N., and support from the Max Planck quity and institutions may only facilitate learning the rules and norms Institute for Evolutionary Anthropology (MPI) to R.M., A.P., and B.G.P. for successfully navigating social life (cf. Brodbeck et al., 2013; Cohn J.H. thanks CIFAR. We thank Adam Barnett, Nicholas Chan, Radek et al., 2014; Cronk, 2007; Gerkey, 2013; Lesorogol, 2007; Smaldino, Kundt, Eva Kundtová Klocová, Tiffany Lai, Peter Manǒ, and Ted 2014), thus making them more salient in situ. While our results suggest Slingerland. We also thank the Department of Human Behavior, that the composition of mental models predicts cooperative behavior Ecology and Culture at MPI for their feedback, particularly Bret Beheim that transcends what might otherwise be parochial boundaries (e.g., for his help with the by-site plots in the Supplementary data. We also Hruschka et al., 2014; Pisor & Gurven, 2016), examining when, and thank the reviewers and Dan Fessler for their critical feedback. where, and to whom participants claim these moral prescriptions apply would be a logical next step for future inquiry (cf. Fessler et al., 2015). Appendix A. Supplementary data

Supplementary materials including data, R scripts, and methods 6.4. Evolution of moral systems protocols to this article can be found online in https://github.com/ bgpurzycki/Moral-Models-Moral-Behavior and at Purzycki et al. Social systems have long been held to structure human interactions, (2016b, 2017). Supplementary data to this article can be found online but their mechanics are rarely detailed with empirical data. We as- at https://doi.org/10.1016/j.evolhumbehav.2018.04.004. sessed some components of moral systems here by focusing on in- dividual models, local culture, and their contribution to honest beha- References vior between members of disparate communities. Like any social system, moral systems are the aggregate output of the complex inter- Alexander, R. D. (1987). The biology of moral systems. New Brunswick: Aldine Transaction. actions between deeper cognitive adaptations and our socioecological Apicella, C. L., Marlowe, F. W., Fowler, J. H., & Christakis, N. A. (2012). Social networks environments (Barrett, 2014; Kiper & Sosis, 2014). Further considera- and cooperation in hunter-gatherers. Nature, 481(7382), 497. tion of deeper psychological systems (Cosmides & Tooby, 2005; Curry, Ashton, M. C., Lee, K., & de Vries, R. E. (2014). The hexaco honesty-humility, agree- ableness, and emotionality factors a review of research and theory. Personality and 2016; Graham et al., 2011) and their relationship to culture and in- Social Psychology Review, 18(2), 139–152. stitutions (Gerkey, 2013; Stagnaro et al., 2017) is necessary to further Barrett, H. C. (2014). The shape of thought: How mental adaptations evolve. Oxford: Oxford assess the dynamism between moral systems and human cooperation. University Press. As is true with all cross-sectional and correlational research, we Barrett, H. C., Bolyanatz, A., Crittenden, A. N., Fessler, D. M., Fitzpatrick, S., Gurven, M.,, et al. ... (2016). Small-scale societies exhibit fundamental variation in the role of cannot satisfactorily explain variation or account for the causal links intentions in moral judgment. Proceedings of the National Academy of Sciences, and feedback between facets of moral systems. The variability we see in 113(17), 4688–4693. moral culture may reflect variation in the challenges people collectively Baumard, N. (2016). The origin of fairness. Oxford: Oxford University Press. Baumard, N., André, J.-B., & Sperber, D. (2013). A mutualistic approach to morality: The face together, but whether or not it contributes to the resolution of evolution of fairness by partner choice. Behavioral and Brain Sciences, 36(1), 59–78. those challenges remains an open question (Alexander, 1987; Curry, Baumard, N., & Sperber, D. (2010). Weird people, yes, but also weird experiments. – – 2016; Greene, 2013; Haidt & Kesebir, 2010). As is the case for most Behavioral and Brain Sciences, 33(2 3), 84 85. ffi Bernard, H. R. (2011). Research methods in anthropology. Lanham: AltaMira Press. traits, uncovering the genesis of moral systems is a di cult-if not im- Blasi, A. (1980). Bridging moral cognition and moral action: A critical review of the lit- possible-task, but longitudinal research on the topic would be better erature. Psychological Bulletin, 88(1), 1–45. able to address the links between cognitive adaptations, culture, and Boehm, C. (1979). Some problems with altruism in the search for moral universals. Behavioral Science, 24(1), 15–24. environment by tracking individual- and group-level moral models and Boehm, C. (1980). Exposing the moral self in Montenegro: The use of natural definitions behavior, including how they change, from which sources they appear to keep ethnography descriptive. American Ethnologist, 7(1), 1–26. to develop, and the forces at work in their selective retention. Bowles, S., & Gintis, H. (2003). Origins of human cooperation. Genetic and Cultural Evolution of Cooperation, 2003, 429–443. Boyd, R. (2018). Adifferent kind of animal: How culture transformed our species. Princeton: Princeton University Press. Author contributions Boyd, R., & Richerson, P. J. (1988). Culture and the evolutionary process. Chicago: University of Chicago Press. B.G.P. conceived, initiated, and managed this project, wrote the Boyd, R., & Richerson, P. J. (1992). Punishment allows the evolution of cooperation (or – bulk of the manuscript, supplementary data, and R code, performed anything else) in sizable groups. Ethology and sociobiology, 13(3), 171 195. Boyd, R., & Richerson, P. J. (2009). Culture and the evolution of human cooperation. analysis and contributed to model development. A.P. contributed to Philosophical Transactions of the Royal Society of London B: Biological Sciences, writing the manuscript, supplementary data, and R code. R.M. con- 364(1533), 3281–3288. tributed to main model development, its description, corollary R code, Brodbeck, F. C., Kugler, K. G., Reif, J. A., & Maier, M. A. (2013). Morals matter in eco- nomic games. PloS one, 8(12), e81558. and group-level plots. C.A., Q.D.A., E.C., R.A.M., B.G.P., A.K.W., and Brown, D. (1991). Human universals. Boston: McGraw-Hill. D.X. collected data and contributed to protocol design with J.H. and Buchtel, E. E., Guan, Y., Peng, Q., Su, Y., Sang, B., Chen, S. X., & Bond, M. H. (2015). A.N. All authors provided feedback. Immorality east and west: Are immoral behaviors especially harmful, or especially

499 B.G. Purzycki et al. Evolution and Human Behavior 39 (2018) 490–501

uncivilized? Personality and Social Psychology Bulletin, 41(10), 1382–1394. Economic Behavior & Organization, 93, 328–336. Caton, C. E. (1963). In what sense and why ‘ought’–judgements are universalizable. The Johnson, D. (2016). God is watching you: How the fear of god makes us human. Oxford: Philosophical Quarterly, 13(50), 48–55. Oxford University Press. Chudek, M., & Henrich, J. (2011). Culture–gene coevolution, norm-psychology and the Kant, I. (1997 [1785]). Groundwork of the metaphysics of morals. Cambridge: Cambri dge emergence of human prosociality. Trends in Cognitive Sciences, 15(5), 218–226. University Press. Cohn, A., Fehr, E., & Maréchal, M. A. (2014). Business culture and dishonesty in the Kiper, J., & Sosis, R. (2014). Moral intuitions and the religious system: An adaptationist banking industry. Nature, 516(7529), 86–89. account. Philosophy, Theology and the Sciences, 1(2), 172–199. Cone, J., & Rand, D. G. (2014). Time pressure increases cooperation in competitively Kurzban, R., Dukes, A., & Weeden, J. (2010). Sex, drugs and moral goals: Reproductive framed social dilemmas. PLOS One, 9(12), e115756. strategies and views about recreational drugs. Proceedings of the Royal Society of Cosmides, L., & Tooby, J. (2005). Neurocognitive adaptations designed for social ex- London B: Biological Sciences, 277(1699), 3501–3508. change. The handbook of evolutionary psychology, 584–627. Lesorogol, C. K. (2007). Bringing norms in: The role of context in experimental dictator Cronk, L. (1994). Evolutionary theories of morality and the manipulative use of signals. games. Current Anthropology, 48(6), 920–926. Zygon, 29(1), 81–101. Lotz, S. (2015). Spontaneous giving under structural inequality: Intuition promotes co- Cronk, L. (2007). The influence of cultural framing on play in the trust game: A Maasai operation in asymmetric social dilemmas. PLoS One, 10(7), e0131562. example. Evolution and Human Behavior, 28(5), 352–358. Machery, E., & Mallon, R. (2010). Evolution of morality. The Moral Psychology Curry, O. S. (2016). Morality as cooperation: A problem-centred approach. In T. K. Handbook, 3. Shackelford, & R. D. Hansen (Eds.). The evolution of morality (pp. 27–51). Springer. McElreath, R. (2016). Statistical rethinking: A Bayesian course with examples in r and stan. Curry, O.S., Chesters, M.J., Van Lissa, C.J., n.d. Mapping morality with a compass: Testing New York: CRC Press. the theory of ‘morality as cooperation’ with a new questionnaire. cited with per- McElreath, R. (2017). rethinking: Statistical rethinking book package R package version 1.63. mission. https://osf.io/x7dg3/. McNamara, R. A., Norenzayan, A., & Henrich, J. (2016). Supernatural punishment, in- D’Andrade, R. G. (2006). Commentary on Searle's social ontology: Some basic principles group biases, and material insecurity: Experiments and ethnography from Yasawa, culture and institutions. Anthropological Theory, 6(1), 30–39. Fiji. Religion, Brain & Behavior, 6(1), 34–55. D’Andrade, R. G. (1981). The cultural part of cognition. Cognitive Science, 5(3), 179–195. Mizzoni, J. (2009). ‘The social instincts naturally lead to the golden rule’: The ethics of Darwin, C. (1871). The descent of man. Princeton, N.J.: Penguin Classics. Charles Darwin. Teorema: Revista Internacional de Filosofía, 123–133. Dawkins, R. (2016). The god delusion. New York: Random House. Nezlek, J. B. (2010). Multilevel modeling and cross-cultural research. Cross-cultural research De Waal, F. (2013). The bonobo and the atheist. New York: W.W. Norton and Company. methods in psychology. Oxford. Ensminger, J., & Henrich, J. (2014). Experimenting with social norms: Fairness and pun- Norris, P., & Inglehart, R. (2012). Sacred and secular: Religion and politics worldwide. ishment in cross-cultural perspective. Russell Sage Foundation. Cambridge: Cambridge University Press. Fehr, E., Fischbacher, U., & Gächter, S. (2002). Strong reciprocity, human cooperation, North, D. C. (1991). Institutions. Journal of Economic Perspectives, 5(1), 97–112. and the enforcement of social norms. Human Nature, 13(1), 1–25. Perry, S. L. (2017). Not practicing what you preach: Religion and incongruence between Fessler, D. M. T., Barrett, H. C., Kanovsky, M., Stich, S., Holbrook, C., Henrich, J., ... pornography beliefs and usage. The Journal of Sex Research, 1–12. Laurence, S. (2015). Moral parochialism and contextual contingency across seven Pinheiro, J. C., & Bates, D. M. (2000). Linear mixed-effects models: Basic concepts and societies. In: Proc. R. Soc. B, 282(1813), 20150907. examples. Mixed-Effects Models in S and S-Plus, 3–56. Fischbacher, U., & Föllmi-Heusi, F. (2013). Lies in disguise—An experimental study on Pisor, A. C., & Gurven, M. (2016). Risk buffering and resource access shape valuation of cheating. Journal of the European Economic Association, 11(3), 525–547. out-group strangers. Scientific Reports, 6. Gelfand, M. J., Raver, J. L., Nishii, L., Leslie, L. M., Lun, J., Lim, B. C.,, et al. ... (2011). Purzycki, B. G. (2011). Tyvan cher eezi and the socioecological constraints of supernatural Differences between tight and loose cultures: A 33-nation study. Science, 332(6033), agents' minds. Religion, Brain & Behavior, 1(1), 31–45. 1100–1104. Purzycki, B. G. (2016). The evolution of gods' minds in the Tyva Republic. Current Gelman, A. (2006). Multilevel (hierarchical) modeling: What it can and cannot do. Anthropology, 57(S13), S88–S104. Technometrics, 48(3), 432–435. Purzycki, B. G., Apicella, C., Atkinson, Q. D., Cohen, E., McNamara, R. A., Willard, A. K., Gerkey, D. (2013). Cooperation in context: Public goods games and post-Soviet collectives ... Henrich, J. (2016a). Moralistic gods, supernatural punishment and the expansion in Kamchatka, Russia. Current Anthropology, 54(2), 144–176. of human sociality. Nature, 530(7590), 327–330. Gervais, M. M. (2017). Rich economic games for networked relationships and commu- Purzycki, B. G., Apicella, C., Atkinson, Q. D., Cohen, E., McNamara, R. A., Willard, A. K., nities: Development and preliminary validation in Yasawa, Fiji. Field Methods, 29(2), ... Henrich, J. (2016b). Cross-cultural dataset for the evolution of religion and mor- 113–129. ality project. Scientific Data, 3, 160099. Graham, J., Haidt, J., Koleva, S., Motyl, M., Iyer, R., Wojcik, S. P., & Ditto, P. H. (2013). Purzycki, B. G., & Jamieson-Lane, A. (2016). AnthroTools: An R package for cross-cultural Moral foundations theory: The pragmatic validity of moral pluralism. Advances in ethnographic data analysis. Cross-Cultural Research, 51(1), 51–74. Experimental Social Psychology, 47,55–130. Purzycki, B., Apicella, C., Atkinson, Q., Cohen, E., Henrich, J., McNamara, R., ... Graham, J., Meindl, P., Koleva, S., Iyer, R., & Johnson, K. M. (2015). When values and Xygalatas, D. (2017). Evolution of religion and morality project dataset (wave 1)http:// behavior conflict: Moral pluralism and intrapersonal moral hypocrisy. Social and dx.doi.org/10.7910/DVN/RT5JTV. Personality Psychology Compass, 9(3), 158–170. Purzycki, B. G., Henrich, J., Apicella, C., Atkinson, Q. D., Baimel, A., Cohen, E., Graham, J., Nosek, B. A., Haidt, J., Iyer, R., Koleva, S., & Ditto, P. H. (2011). Mapping the McNamara, R. A., Willard, A., Xygalatas, D., & Norenzayan, A. (2018). The evolution moral domain. Journal of Personality and Social Psychology, 101(2), 366. of religion and morality: A synthesis of ethnographic and experimental evidence from Greene, J. (2013). Moral tribes: Emotion, reason, and the gap between us and them. New eight societies. Religion, Brain and Behavior, 8(2), 101–132. York: Penguin. R Core Team (2016). R: A language and environment for statistical computing. Vienna, Gurven, M., Von Rueden, C., Massenkoff, M., Kaplan, H., & Lero Vie, M. (2013). How Austria. universal is the big five? Testing the five-factor model of personality variation among Richerson, P., Baldini, R., Bell, A. V., Demps, K., Frost, K., Hillis, V.,, et al. ... (2016). forager–farmers in the Bolivian amazon. Journal of Personality and Social Psychology, Cultural group selection plays an essential role in explaining human cooperation: A 104(2), 354. sketch of the evidence. Behavioral and Brain Sciences, 39, e30. Gurven, M., & Winking, J. (2008). Collective action in action: Prosocial behavior in and Roth, A. E., Prasnikar, V., Okuno-Fujiwara, M., & Zamir, S. (1991). Bargaining and market out of the laboratory. American Anthropologist, 110(2), 179–190. behavior in Jerusalem, Ljubljana, Pittsburgh, and Tokyo: An experimental study. The Gurven, M., Zanolini, A., & Schniter, E. (2008). Culture sometimes matters: Intra-cultural American Economic Review, 1068–1095. variation in pro-social behavior among Tsimane amerindians. Journal of Economic Rozin, P. (1999). The process of moralization. Psychological Science, 10(3), 218–221. Behavior & Organization, 67(3–4), 587–607. Schwartz, S. H. (2007). Universalism values and the inclusiveness of our moral universe. Haidt, J. (2001). The emotional dog and its rational tail: A social intuitionist approach to Journal of Cross-Cultural Psychology, 38(6), 711–728. moral judgment. Psychological review, 108(4), 814–834. Searle, J. R. (1995). The construction of social reality. New York: Simon and Schuster. Haidt, J., & Kesebir, S. (2010). Morality. In S. Fiske, & D. Gilbert (Eds.). Handbook of social Shweder, R. A., Much, N. C., Mahapatra, M., & Park, L. (1997). The “big three” of mor- psychology (pp. 797–832). Hoboken: Wiley. ality (autonomy, community, divinity) and the “big three” explanations of suffering. Handwerker, W. P. (2001). Quick ethnography. Lanham: Altamira Press. In A. M. Brandt, & P. Rozin (Eds.). Morality and health (pp. 119– 169). Taylor & Henrich, J. (2000). Does culture matter in economic behavior? Ultimatum game bar- Frances/Routledge. gaining among the Machiguenga of the Peruvian amazon. The American Economic Smaldino, P. E. (2014). The cultural evolution of emergent group-level traits. Behavioral Review, 90(4), 973–979. and Brain Sciences, 37(3), 243–254. Henrich, J., Boyd, R., Bowles, S., Camerer, C., Fehr, E., & Gintis, H. (Eds.). (2004). Smith, C. E., Blake, P. R., & Harris, P. L. (2013). I should but I wont: Why young children Foundations of human sociality: Economic experiments and ethnographic evidence from endorse norms of fair sharing but do not follow them. PLOS One, 8(3), e59510. fifteen small-scale societies. New York: Oxford University Press. Smith, J. J. (1993). Using Anthropac 3.5 and a spreadsheet to compute a free-list salience Henrich, J., & Henrich, N. (2014). Fairness without punishment: Behavioral experiments index. Cultural Anthropology Methods, 5(3), 1–3. in the Yasawa islands, Fiji. In J. Ensminger, & J. Henrich (Eds.). Experimenting with Smith, J. J., & Borgatti, S. P. (1997). Salience counts—And so does accuracy: Correcting social norms: Fairness and punishment in cross-cultural perspective (pp. 225–258). SAGE. and updating a measure for free-list-item salience. Journal of Linguistic Anthropology, Hofmann, W., Wisneski, D. C., Brandt, M. J., & Skitka, L. J. (2014). Morality in everyday 7(2), 208–209. life. Science, 345(6202), 1340–1343. Smith, J. J., Furbee, L., Maynard, K., Quick, S., & Ross, L. (1995). Salience counts: A Hruschka, D., Efferson, C., Jiang, T., Falletta-Cowden, A., Sigurdsson, S., McNamara, R., domain analysis of english color terms. Journal of Linguistic Anthropology, 5(2), ... Henrich, J. (2014). Impartial institutions, pathogen stress and the expanding social 203–216. network. Human Nature, 25(4), 567–579. Smith, K. D., Smith, S. T., & Christopher, J. C. (2007). What defines the good person? Jamieson-Lane, A., & Purzycki, B. G. (2016). AnthroTools: Some custom tools for anthro- Cross-cultural comparisons of experts' models with lay prototypes. Journal of Cross- pology R package version 0.8. Cultural Psychology, 38(3), 333–360. Jiang, T. (2013). Cheating in mind games: The subtlety of rules matters. Journal of Sperber, D. (1996). Explaining culture: A naturalistic approach. Blackwell Oxford.

500 B.G. Purzycki et al. Evolution and Human Behavior 39 (2018) 490–501

Sperber, D., & Baumard, N. (2012). Moral reputation: An evolutionary and cognitive Trivers, R. L. (1971). The evolution of reciprocal altruism. The Quarterly Review of Biology, perspective. Mind & Language, 27(5), 495–518. 46(1), 35–57. Sripada, C. S., & Stich, S. (2006). A framework for the psychology of norms. In P. Turiel, E. (1983). The development of social knowledge: Morality and convention. Cambridge: Carruthers, S. Laurence, & S. Stich (Vol. Eds.), The innate mind: Culture and cognition. Cambridge University Press. Vol.2. The innate mind: Culture and cognition (pp. 280–301). Oxford: Oxford University Turiel, E. (2006). Thought, emotions, and social interactional processes in moral devel- Press. opment. In M. Killen, & J. Smetana (Eds.). Handbook of moral development (pp. 7–35). Stagnaro, M. N., Arechar, A. A., & Rand, D. G. (2017). From good institutions to generous Lawrence Erlbaum Associates, Publishers. citizens: Top-down incentives to cooperate promote subsequent prosociality but not Wierzbicka, A. (1994). Semantic primitives across languages: A critical review. Semantic norm enforcement. Cognition, 167, 212–254. and lexical universals: Theory and empirical findings, 445–500. Stan Development Team (2017). RStan: The R interface to Stanhttp://mc-stan.org/ R Wiessner, P. (2009). Experimental games and games of life among the Ju/’hoan bushmen. package version 2.16.2. Current Anthropology, 50(1), 133–138. Tomasello, M., & Vaish, A. (2013). Origins of human cooperation and morality. Annual Winking, J., & Mizer, N. (2013). Natural-field dictator game shows no altruistic giving. Review of Psychology, 64, 231–255. Evolution and Human Behavior, 34(4), 288–293.

501